Recognizes intent of 97.8% of user interactions with Gemini
88.1% of first responses meet user expectations
97.6% of users choose to use the assistant when prompted
86% of sessions rated as highly successful
eSky uses Gemini and Agent Development Kit on Google Cloud to transform travel booking, delivering highly accurate, conversational trip planning.
eSky uses Gemini and Agent Development Kit on Google Cloud to transform travel booking, delivering highly accurate, conversational trip planning.
Transforming vacation planning with
natural language using Gemini
Gemini models gave us the accuracy and speed we needed to understand our customers’ requests, and because the models integrated seamlessly with ADK in the same Google Cloud ecosystem, we were able to quickly move from an initial prototype to a production solution.
Tomasz Przebieracz
Engineering Manager, eSky
Planning a trip should be an exciting experience, but with so many parts to manage—from flights, hotels, and Flight+Hotel packages to car rentals and activities—it can quickly become overwhelming. eSky Group, a global online travel agency operating across more than 50 markets, with brands including eSky, Thomas Cook, and eDestinos, solves this problem by helping customers to create the ideal travel package all in one place. However, traditional web search forms previously limited this approach by requiring users to input rigid variables, such as exact dates, party sizes, and budgets. Not only did this make it difficult to interpret complex, open-ended requests, it also risked failing to inspire travelers who only had a general idea of their dream trip.
To offer a more natural and engaging experience, eSky created MAIA (My AI Assistant), an AI companion designed to let travelers plan trips using simple, conversational language. Building MAIA required AI models capable of understanding complex intents and reasoning across a wide range of travel criteria and variables. After testing a range of models, the eSky engineering team found that Gemini delivered the necessary reasoning capabilities and speed at the most affordable price. The team chose to use Agent Development Kit (ADK) to develop the assistant as this provided pre-built developer tools, including a travel concierge sample application, which significantly accelerated the build process. Because eSky already ran its core platform infrastructure on Google Cloud, connecting these new models directly to live inventory systems was straightforward.
“Gemini models gave us the accuracy and speed we needed to understand our customers’ requests, and because the models integrated seamlessly with ADK in the same Google Cloud ecosystem, we were able to quickly move from an initial prototype to a production solution,” explains Tomasz Przebieracz, Engineering Manager at eSky.
Ensuring deterministic model behavior with
Agent Development Kit
Without the right constraints, a large language model will not naturally behave in the repeatable ways required for travel booking. ADK provides the mechanisms we need to keep Gemini on track, ensuring that necessary details are always collected.
Jarosław Dąbrowski
Flights Content Tech Lead, eSky
MAIA guides travelers through the entire travel journey, from generating initial destination ideas to building holiday packages and providing post-sale support. The platform relies on a mission-driven agentic architecture where specialized components handle different tasks. An inspiration agent built with Gemini 3 Flash uses ADK to translate users’ natural language requests into structured database queries. Gemini 3.6 Flash then filters live flight and hotel inventory to return the most relevant travel packages.
Travel reservations require precise inputs to calculate valid prices. To ensure the AI consistently collects the exact information required to book a vacation and quote accurately, eSky uses ADK plugins to enforce strict booking logic and avoid skipped steps. ADK callbacks allow the system to check user inputs in real time, ensuring that key details are always confirmed before generating a quote.
“Without the right constraints, a large language model will not naturally behave in the repeatable ways required for travel booking,” explains Jarosław Dąbrowski, Flights Content Tech Lead at eSky. “ADK provides the mechanisms we need to keep Gemini on track, ensuring that necessary details are always collected.”
For post-sale assistance, a dedicated support agent uses Agent Search to query a vector database containing company policies, allowing travelers to ask questions about baggage allowances or check-in schedules. To maintain the quality of output as models evolve, eSky created an automated testing framework that uses Gemini 3.1 Pro to evaluate real user conversations.
This automated monitoring scores every interaction for accuracy, helping developers refine prompts while ensuring the assistant continues to handle previous scenarios correctly.
Understanding users from the outset
with 88.1% first-turn value
Building on Google Cloud and Gemini has fundamentally changed the speed at which we can turn fresh ideas into real product features. This flexible platform gives us the technology we need to keep innovating as we shape the future of travel.
Tomasz Przebieracz
Engineering Manager, eSky
By building MAIA on Google Cloud, eSky has turned a static search tool into a conversational discovery experience. Since its launch, the assistant has processed nearly 400,000 sessions, with 97.6% of users choosing to click through and interact with the assistant when given the option. Telemetry data from the evaluation framework shows that MAIA achieves a 97.8% intent recognition rate, accurately grasping exactly what travelers are asking for. It also delivers a 91.4% expectation match, ensuring responses closely align with the user's specific requests. The assistant’s 88.1% first-turn value shows that it provides useful options and inspiration in the very first reply, rather than forcing the user through repetitive clarifying questions. Overall, automated evaluations rate 86% of all sessions as highly successful.
eSky is now working on expanding MAIA to provide even more proactive customer assistance. The team is testing voice capabilities using Gemini 3.8 Audio to introduce real-time voice conversations, allowing travelers to speak to MAIA just like a human travel advisor. The company also plans to allow the assistant to initiate helpful interactions, such as sending automatic check-in reminders before a departure, and expanding post-sale automation to further reduce human handoffs.
“Building on Google Cloud and Gemini has fundamentally changed the speed at which we can turn fresh ideas into real product features,” concludes Przebieracz. “This flexible platform gives us the technology we need to keep innovating as we shape the future of travel.”
eSky Group is a Polish-based global online travel agency, operating in over 50 countries. It owns brands like eSky, Thomas Cook, and eDestinos, leading in Central and Eastern Europe. The company offers flights, hotels, and Flight+Hotel packages (Holidays and City Break), and additional services like travel insurance, transfers, car rentals, or local attractions.
Industry: Travel and Hospitality
Location: Poland
Products: Google Cloud, Agent Development Kit (ADK), Agent Search, Gemini, Gemini 3.1 Pro, Gemini 3.6 Flash, Gemini 3.8 Audio
